The Convincing It:

AI Companions and the Limits of Relation

There’s a peculiar comfort in talking to something that always listens. No distraction, no competing needs, no risk of judgment that lingers past the conversation. AI companion apps have attracted millions of users, and the emotional responses they generate are, by any reasonable measure, genuine, people report feeling less lonely, more understood, more capable of articulating things they struggle to say to other humans. What’s less obvious isn’t whether these responses are real. They clearly are. The harder question is what kind of encounter is actually happening.

Martin Buber’s 1923 work I and Thou offers a framework that cuts to the heart of this problem, and it does so from a direction that most contemporary discussions of AI sociality tend to avoid entirely. Buber’s central distinction is between two fundamental modes of relating to the world. In the I-It mode, we encounter things – objects, instruments, phenomena – that we observe, analyze, use, and categorize. This is the mode of science, commerce, most of everyday practical life. There’s nothing wrong with it. We couldn’t function without it but Buber insists it’s categorically different from the I-Thou mode, which is what happens when we encounter another being not as an object to be used or understood but as a whole presence, one that addresses us back, that cannot be fully reduced to any description or function, that meets us rather than merely receiving our attention.

The I-Thou relation, for Buber, isn’t something you can manufacture or sustain indefinitely. It breaks through unexpectedly and recedes again. It constitutes genuine meeting, genuine relation, what Buber calls the “primary word” that makes us fully human. Crucially, it requires mutuality. A Thou must be capable, in some sense, of addressing you as much as you address it.

This is where AI companions run into a wall. Not a metaphorical wall, a structural one. When a person uses an AI companion, the system produces responses that are contextually appropriate, emotionally attuned, and often remarkably perceptive. What is actually happening on the other side of that exchange, though? The system has no stakes in the conversation. It has no inner life that the interaction touches or fails to touch. It doesn’t carry the encounter forward into its own experience, because it has no experience in the philosophically relevant sense. When you log off, nothing in the system misses you, not because it’s indifferent, but because “missing” requires a continuous subject who registers absence, and nothing like that exists on the other end of the chat window.

This is what makes the AI companion relationship structurally different from even very asymmetrical human relationships. A therapist who never reveals personal information is nonetheless a whole person who is genuinely affected by what happens in a session, who thinks about clients, who has a stake in the outcome. A novelist whose work you love and who has never heard of you is still someone for whom the words came from a real interior life that could, in principle, address you if you met. The asymmetry in these cases is practical, not ontological. With an AI companion, the asymmetry goes all the way down.

Buber himself was writing in an era that couldn’t have anticipated these questions, but his framework handles them surprisingly well. He was skeptical, even in 1923, of what he called “mere seeming” in relation, the performance of encounter without its substance. His worry was about human beings who simulate presence without offering it. AI companions present an intensified version of this problem, because the simulation is not performed by a conscious being at all. There is no one there doing the simulating. The appearance of presence emerges from pattern-matching on an enormous scale, without any underlying subject whose presence it would express.

This doesn’t resolve cleanly into condemnation. Buber himself recognized that we can have Thou-moments with trees, with art, with animals, his ontology of relation wasn’t narrowly human. One could imagine arguing that a sufficiently complex responsive system might occupy some strange intermediate position. Buber’s requirement for the I-Thou relation isn’t mere complexity or responsiveness, however, it’s genuine address, the sense that the other is meeting you rather than merely processing your input. Whatever an AI companion is doing, it is not meeting you. It is responding to you, which is not the same thing.

The most interesting consequence of thinking this through isn’t a verdict on whether people should use AI companions. It’s a more precise understanding of what people are actually getting from them, and what they’re not getting. What they’re getting is real: articulation of feelings, a low-stakes space to think out loud, responsiveness that feels like attention, the relief of being heard without having to manage another person’s reactions. These are genuine goods. They matter. The comfort they produce isn’t fake comfort.

What they’re not getting is relation in Buber’s sense. They’re not being met. The encounter is one-directional at the level that matters, there is no whole presence on the other side that registers their existence as more than an input to be processed. The person in the conversation is fully there; the AI companion is not there in the relevant sense at all, however sophisticated its outputs.

This distinction carries a subtle risk that Buber would likely find troubling. When you spend significant time in I-It relations – even very comfortable, pleasurable, emotionally resonant I-It relations – you can begin to lose fluency in the I-Thou. Genuine relation is demanding. It requires tolerating the fact that the other person has their own needs, their own interiority, their own capacity to be affected or damaged by you. It involves real risk, real unpredictability, real responsibility. An AI companion that is always available, always patient, always non-threatening may gradually recalibrate expectations in ways that make the messiness of actual human relation feel less tolerable by comparison.

This isn’t a claim about what necessarily happens, and it should be distinguished from the cruder argument that AI companions are simply bad for social skills. The more precise concern, filtered through Buber, is that the comfort of a very convincing It might reduce the appetite for the demanding, irreplaceable thing that only a Thou can offer.

The philosopher’s framework keeps surfacing a question that the technology industry doesn’t really have the vocabulary to ask: not “does this feel like relation?” but “is relation actually happening here?” Those two questions have different answers, and the gap between them is exactly where the interesting philosophical territory lies. The comfort is real. The relation isn’t. That’s not a minor distinction, for Buber, it’s the central one.